Transcript: Ousted OpenAI board member on AI safety concerns

This is an audio transcript of the FT News Briefing podcast episode: ‘Ousted OpenAI board member on AI safety concerns’

Persis Love
Good morning from the Financial Times. Today is Thursday, November 23rd. And this is your FT News Briefing.

Sam Altman is back at OpenAI. The US suspects India was involved in a Sikh assassination plot. And the UK government outlines its spending plans.

Lucy Fisher
The government want us to be talking about cutting taxes, but nonetheless, the taxes in the UK are heading to a postwar high.

Persis Love
I’m Persis Love, in for Marc Filippino. And here’s the news you need to start your day.

[MUSIC PLAYING]

The latest in the OpenAI boardroom drama: Sam Altman has been reinstated as CEO and some of the board members who tried to oust him last week are losing their seats. But there are still questions about why they tried to push out Altman. He’s a key figure in the latest AI boom. Helen Toner was one of the board members who moved against the chief executive. The FT’s Madhumita Murgia interviewed her last month before the boardroom drama kicked off. Madhu joins us now to tell us more. Hi, Madhu.

Madhumita Murgia
Hi, there.

Persis Love
So bring us up to date then, Madhu. How did Sam Altman end up back at the top?

Madhumita Murgia
So it’s been a real rollercoaster since Friday. But really what happened over the last few days is after Sam Altman was ousted by the board, there was this groundswell of support from investors, but also from employees at OpenAI who really lobbied to bring him back. Now, we don’t quite know why he was removed in the first place. There are some ideas about it, but the board members have said is that they felt that Sam hadn’t been transparent with them, and they felt that he could quite be relied upon or trusted. One of these board members was Helen Toner. She works at a think-tank out of Georgetown University and focuses on issues of AI safety, particularly in relation to national security and looking at China and geopolitics also. I spoke to Helen before all of this drama began, and it gives us a little bit of a glimpse into her thinking and her perspective. One of the things she did mention to me that she cared a lot about was AI safety.

Persis Love
And what are the specific safety issues that she mentioned to you?

Madhumita Murgia
She talked about a few different examples of things that concerned her. She talked about how generative AI, particularly symbolised by ChatGPT, had been getting better and better. She talked also about misinformation around elections using generative AI and things like hacking becoming a lot easier if AI can code. 

Helen Toner
ChatGPT, one of the things that’s kind of most useful for, one of the most sort of intellectually demanding tasks that it’s useful for right now, is programming. And you know, right now, again, it’s not really sort of capable or powerful enough to be particularly dangerous. But if you have a system that is really good at programming and, you know, if the researchers working on this get their way, then ChatGPT will get much better at programming over the next few years. A system that’s very good at programming is also very good at hacking, right? So it could potentially enable a much broader group of people to be able to carry out sophisticated cyber attacks, perhaps including cyber attacks on critical infrastructure or other really significant targets.

Madhumita Murgia
But her biggest concern really was around how the progress of AI, the development, has been really steady, and it shows no signs of slowing down. And she wasn’t sure that the people building these systems, which includes OpenAI of course and Altman, were being responsible about how it was evolving.

Helen Toner
Fundamentally, we really don’t understand what we’re building. We don’t understand how it does what it does. We don’t understand how to constrain it. And I think that means that we need to have a lot of humility about how confident we can be.

Madhumita Murgia
The point here is that nobody really understands how generative AI works on the inside, and that, she thinks, is a big problem.

Persis Love
So again, this interview you had with Helen Toner happened before the drama at OpenAI, and we don’t know exactly what her reasoning was — but did she say anything that might give some insight into what she was thinking?

Madhumita Murgia
We did briefly touch on her role as a board member at OpenAI during our interview, and she seemed quite worried about oversight and who was really able to have the power to oversee companies like OpenAI.

Helen Toner
I think the people at the top of top AI companies generally take the risks pretty seriously. You know, depends on the company, depends on the person. I think for the most part, they are wanting to do the right thing. At the same time, you know, they’re obviously the ones building these systems. They’re the ones who potentially stand to profit from them. So I think it’s really important to make sure that there is outside oversight, not just by the boards of the companies, but also by regulators and by the broader public. So even if their hearts are in the right place, we shouldn’t rely on that as our primary way of ensuring they do the right thing.

Madhumita Murgia
So she was quite pointed here, really, telling us that there needed to be better oversight because there were these conflicts of interest within companies who were trying to make money while also trying to safeguard this powerful technology.

Persis Love
And finally, Madhu, what will you be watching for as this story evolves?

Madhumita Murgia
So I don’t think this is going to be the end of it. Not at OpenAI necessarily, but also more broadly. We’re going to see more of these clashes as experts have really differing views on the future of AI, on safety issues surrounding AI and really on what needs to be done to make sure we can control this really fast-evolving technology.

Helen Toner
Something that’s really important to know about the field of AI and AI scientists right now is that they disagree wildly about all kinds of things. They disagree about how impressive current systems are. They disagree about how impressive the systems that we’ll have in five or 10 years are. And so that kind of puts us in this tricky position where I think it is reasonable for any sort of onlooker, for a policymaker, for a member of the general public to be very uncertain about what we’re likely to experience. But that uncertainty can’t paralyse us. It has to be uncertainty that catalyses action to manage the potential risks.

Madhumita Murgia
So she certainly did take action, and we may well see more of this throughout the industry and this AI ecosystem going forward.

Persis Love
Thanks, Madhu.

Madhumita Murgia
Thank you for having me.

[MUSIC PLAYING]

Persis Love
Madhumita Murgia is the FT’s artificial intelligence editor and the co-host of the Tech Tonic podcast. Tech Tonic’s latest series on the pursuit of artificial general intelligence is out now.

[MUSIC PLAYING]

US authorities have thwarted a plot to kill a Sikh separatist leader on American soil. And Washington has issued a warning to the Indian government over concerns it was involved in the assassination attempt. The target was a US-Canadian citizen and leader of a group campaigning for an independent Sikh state. The Indian government has said it is investigating Washington’s claim since the FT broke the story. All of this comes as India has accused Canada, the UK and other countries of harbouring Sikh separatists, who it frequently brands as terrorists.

[MUSIC PLAYING]

UK chancellor Jeremy Hunt delivered his Autumn Statement yesterday, spelling out the Conservative party’s spending plans for the UK economy. Hunt headlined his budget as the biggest tax cuts in British history, but the country’s fiscal watchdog took a very different view. Here to explain is the FT’s Whitehall editor Lucy Fisher. Hey, Lucy.

Lucy Fisher
Hi, Persis. 

Persis Love
So tell us, what are the main takeaways from Hunt’s statement?

Lucy Fisher
Well, the two big measures that the government want us to be talking about are tax cuts. The first is a key tax cut for workers in the UK worth about £450 a year on average to those in employment. The second is a big tax cut for business, making the full expensing capital allowance regime permanent. That is a tax break for businesses that invest in plants and machinery. So the government want us to be talking about cutting taxes, but nonetheless, the taxes in the UK are heading to a postwar high. I’ll also just say a very quick word on the economic picture, which looks pretty bleak. The fiscal watchdog today has slashed the UK’s growth forecast.

Persis Love
So how does that work? How can Hunt be calling this statement tax cutting when actually the tax burden is increasing?

Lucy Fisher
Great question. It’s because while they are cutting two very eye-catching taxes, many of the tax thresholds in the UK remain frozen. So during a time of inflation, it’s still dragging people into higher bands, and that’s why overall taxes are rising.

Persis Love
The UK faces a general election next year, and the Conservative party is trailing in the polls. Will this budget do anything to help their chances of re-election?

Lucy Fisher
Well, the ruling Conservative party are trailing by an average of 20 points, so it’s a huge gap for them to try and make up. Interestingly, there’s been a lot of rancour among the ruling Conservative party MPs about the direction they’re going in under the prime minister. Much of the reaction I’ve picked up today has been positive. MPs on the right wing of the party are very happy to see tax cuts for business and for individuals. On the more centrist wing, there’s been a certain degree of praise for Jeremy Hunt, the chancellor, uprating benefits, uprating the state pension, raising the minimum wage and also doing more to help people who rely on housing allowance. However, is this enough to move the dial? Many MPs are still pretty sceptical. One other thing to highlight is lots of MPs have been really interested in this tax cut for workers. It’s being expedited to be introduced in January rather than the start of the next tax year. And that’s got a lot of people thinking that maybe the government is bringing forward this move because it wants voters to feel richer in their pockets earlier in the year because they might want to call an earlier election. The government has to call a general election by January the 28th, 2025. A lot of the thinking and speculation is that it might be in the autumn. But this move might give them manoeuvre room to go earlier in the spring.

Persis Love
Lucy Fisher is the FT’s Whitehall editor. Thanks for joining me, Lucy.

Lucy Fisher
Thanks for having me, Persis.

Persis Love
For more details on that Autumn Statement, check out the FT’s Political Fix podcast, out now.

[MUSIC PLAYING]

You can read more on all these stories at ft.com for free when you click on the links in our show notes. This has been your daily FT News Briefing. Make sure you check back tomorrow for the latest business news.